RESEARCH ARTICLE Open Access

Comprehensive transcriptomic analysis ofheat shock proteins in the molecularsubtypes of human breast cancerFelipe C. M. Zoppino*† , Martin E. Guerrero-Gimenez†, Gisela N. Castro and Daniel R. Ciocca

Abstract

Background: Heat Shock Proteins (HSPs), a family of genes with key roles in proteostasis, have been extensivelyassociated with cancer behaviour. However, the HSP family is quite large and many of its members have not beeninvestigated in breast cancer (BRCA), particularly in relation with the current molecular BRCA classification. In thiswork, we performed a comprehensive transcriptomic study of the HSP gene family in BRCA patients from both TheCancer Genome Atlas (TCGA) and the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC)cohorts discriminating the BRCA intrinsic molecular subtypes.

Methods: We examined gene expression levels of 1097 BRCA tissue samples retrieved from TCGA and 1981 samplesof METABRIC, focusing mainly on the HSP family (95 genes). Data were stratified according to the PAM50 geneexpression (Luminal A, Luminal B, HER2, Basal, and Normal-like). Transcriptomic analyses include several statisticalapproaches: differential gene expression, hierarchical clustering and survival analysis.

Results: Of the 20,531 analysed genes we found that in BRCA almost 30% presented deregulated expression (19%upregulated and 10% downregulated), while of the HSP family 25% appeared deregulated (14% upregulatedand 11% downregulated) (|fold change| > 2 comparing BRCA with normal breast tissues). The study revealedthe existence of shared HSP genes deregulated in all subtypes of BRCA while other HSPs were deregulated inspecific subtypes. Many members of the Chaperonin subfamily were found upregulated while three members(BBS10, BBS12 and CCTB6) were found downregulated. HSPC subfamily had moderate increments of transcripts levels.Various genes of the HSP70 subfamily were upregulated; meanwhile, HSPA12A and HSPA12B appeared stronglydownregulated. The strongest downregulation was observed in several HSPB members except for HSPB1. DNAJ membersshowed heterogeneous expression pattern. We found that 23 HSP genes correlated with overall survival and threeHSP-based transcriptional profiles with impact on disease outcome were recognized.

Conclusions: We identified shared and specific HSP genes deregulated in BRCA subtypes. This study allowedthe recognition of HSP genes not previously associated with BRCA and/or any cancer type, and the identification ofthree clinically relevant clusters based on HSPs expression patterns with influence on overall survival.

Keywords: Breast cancer, Heat shock proteins, Differential gene expression, Molecular subtypes, Survival, HSP-Clusts

* Correspondence: mzoppino@yahoo.com†Felipe C. M. Zoppino and Martin E. Guerrero-Gimenez contributed equallyto this work.Laboratory of Oncology, Institute of Medicine and Experimental Biology ofCuyo (IMBECU), National Scientific and Technical Research Council (CONICET),Av. Dr. Ruiz Leal s/n, Parque General San Martín, 5500 Mendoza, Argentina

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Zoppino et al. BMC Cancer (2018) 18:700 https://doi.org/10.1186/s12885-018-4621-1

BackgroundIn worldwide terms, breast cancer (BRCA) has the sec-ond annual incidence (1,670,000 cases) and the fifthmortality rate (522,000 deaths associated) of overall can-cers [1]. Classifications of BRCA have been performedaccording to clinical features, histological characteristics,and presence of steroid and/or growth factor receptors.PAM50 gene expression assay allows the molecular clas-sification of BRCA based on the expression levels of fiftygenes and sorts BRCA into five intrinsic subtypes:Luminal A, Luminal B, HER2-enriched (HER2), Basal-like(Basal) and Normal-like (Normal). This classificationhighly correlates with BRCA biological behaviour and hasclinical use due to its prognostic significance [2, 3]. HeatShock Proteins (HSPs) are ubiquitous in living organismsand their expression is rapidly regulated by stress. Histor-ically they were recognized as proteins induced by heat, al-though it is now known that various types of physiologicaland/or pathological stresses regulate their expression [4].HSP systems are involved in protein quality control [5],degradation pathways (ubiquitin-proteasome system,endoplasmic reticulum associated degradation, autoph-agy), and regulation of apoptosis [5, 6]. The HSPs belongto a family of evolutionarily conserved genes that includes95 genes divided into five subfamilies: 1) type I chapero-nins (HSP10 and HSP60), BBs chaperonins, and type IIchaperonins (CCT genes) which are grouped under theChaperonin subfamily (CHAP); 2) HSP70 (HSPA) andlarge HSP 100–110 kDa (which are all included in theHSP70 family); 3) small HSP 12–43 kDa (HSPB); 4)HSP90 (HSPC); and 5) HSP40 (DNAJ) [7]. The HSPs re-lated systems can be disturbed during oncogenesis allow-ing malignant transformation and/or facilitating rapidsomatic evolution; they have been studied in a wide varietyof cancers, presenting different pro-tumour (stimulatingtumour growth and metastasis) or anti-tumour ac-tions [4, 8]. Currently, HSPs are emerging as moleculartargets in cancer therapy through the interference of theirdiversity of functions in cancer cells by different ap-proaches. In fact, there are clinical trials for various can-cers, including BRCA, using HSP-inhibitor compoundsand other HSP-based strategies [9–11]. The informationgathered from diverse studies regarding the role of theHSPs in different situations associated with cancer fre-quently provides contradictory overviews. HSP genes (andencoded proteins) corresponding to HSPA1A/B, HSPB1,DNAJB1 and HSP90AA1 are the most studied; these havebeen tested in various models (cell culture, biopsies, etc.),nevertheless in the context of BRCA many others HSPshave not been studied yet. Currently, we have not foundspecific studies of the complete HSP gene family in BRCAintegrating the multi-omics platforms available. The par-ticipation and implications of HSPs involved in differentpathways controlling cell growth, differentiation and

apoptosis emphasize the importance for a thorough andcomprehensive study of all members of these genes. Thepurpose of this study is the analysis and integration ofclinical and transcriptomic (RNAseq) data of BRCAtumour samples from TCGA and METABRIC databaseswith emphasis on HSP genes in the five BRCA mo-lecular subtypes. We hypothesize that the results ofthis investigation will generate relevant knowledge ofthe HSPs expression landscape, useful in the genomicand clinical characterization of BRCA.

MethodsData analysesTwo independent datasets were used in this study: 1)The “TCGA assembler” v.1.0.3 [12] package was used toprogrammatically download, from the publicly availableTCGA (http://cancergenome.nih.gov/) dataset of mam-mary adenocarcinoma, level 3 standardized (normalized)and non-standardized (raw counts) mRNA gene expres-sion levels from 1097 tumour samples and 114 normaltissue samples measured using the RNA-Seq technology(RNASeqV2) (May 1, 2015). Available clinical informa-tion corresponding to 1085 patients was obtained usingthe same package and updated with the latest follow-upavailable. Samples were obtained from patients with ini-tial diagnosis of invasive breast adenocarcinoma under-going surgical resection and that had no prior treatmentfor their diseases. Samples were collected between 1988and 2013, disregarding gender, race, histological type,disease stage or other co-morbidities (Additional file 1:Table S1). The tumour sections analysed were required tocontain an average of 60% tumour cell nuclei with lessthan 20% necrosis under TCGA protocol standards. Thetreatments of patients varied according to the standard oftreatment at time of diagnosis and with the inclusion ofpatients under clinical trial protocols. For further informa-tion about biospecimen collection, processing, qualitycontrol and biomarker assessment, please refer to [3] or toTCGA website (http://cancergenome.nih.gov). 2) To valid-ate the HSP clusters detected in the TCGA dataset, theclinical information and the normalized gene expressionlevels of 1981 tumours from patients with breast cancerwere acquired from the METABRIC cohort. [13]. TheMETABRIC database analyses 49,576 transcripts with Illu-mina HT 12 microarray technology and reports patientoverall survival and disease-specific survival. These datawere accessed through Synapse (synapse.sagebase.org, ID:syn1757063, syn1757053 and syn1757055).The analysis workflow is summarized in Additional file 2.

All analyses and graphs were performed using R softwareenvironment unless otherwise specified. This study hasbeen approved by the Bioethical Committee of theMedical School of the National University of Cuyo,Mendoza, Argentina (0029963/2015).

Zoppino et al. BMC Cancer (2018) 18:700 Page 2 of 17

Intrinsic subtype classificationThe expression levels of the PAM50 panel genes fromeach of the 1097 samples from TCGA were used to carryout the intrinsic subtype classification of tumours [2]which was performed using the “Bioclassifier” package,kindly given by Dr. K. Hoadley of the University ofNorth Carolina Chapel Hill and available online. To per-form this task, the normalized expression profile (nor-malized RNA_SeqV2 RSEM) of the 50 specific geneswas used. Many of these genes are strongly related toBRCA behaviour and include ESR1, ERBB2, PGR, andMKI67 among others. To normalize the expressionvalues from each gene the log2 expression levels wereobtained and subsequently the median expression valueof a subset of samples (50% oestrogen receptor positiveand 50% oestrogen receptor negative population definedby immunohistochemistry) was subtracted. Once thesamples were classified, principal component analysis,class to centroid correlation, and hierarchical clusterevaluations were performed to assess the quality and val-idity of the classification (Additional files 3 and 4). Wefound 89 and 100% concordances with previously re-ported classifications by Koboldt [3] and Ciriello [14]respectively (Additional file 1: Table S5). From thetotal samples analysed from the TCGA cohort, we foundfew cases of the Normal-like subtype (only 3.6%), 51.5%were Luminal A, 20% were Luminal B, 17% were Basal,and 7.5% were HER2, which are in agreement with otherstudies [15, 16]. All 1981 METABRIC patients were classi-fied according to the PAM50 classification as describedabove and Normal-like patients were excluded from fur-ther consideration.

Differential gene expression of TCGA samplesTo evaluate differentially expressed genes (DEG) two dif-ferent statistical packages, DESeq2 [17] and EdgeR [18],were chosen due to their demonstrated good perform-ance [19]. In this study, we used raw count expression of20,531 genes from 1211 tissue samples. We groupedsamples according to the subtypes assigned, and theneach group was compared against normal tissue expres-sion profiles using the standard workflow as presentedin: https://www.bioconductor.org/packages/3.3/bioc/vi-gnettes/DESeq2/inst/doc/DESeq2.pdf and https://bio-conductor.org/packages/release/bioc/vignettes/edgeR/inst/doc/edgeRUsersGuide.pdf. In both cases log2 foldchange values were obtained associated with P valuesand False Discovery Rate values (FDR, a modified Pvalue to correct the eventually false positives) by Benja-mini and Hochberg method [20]. Results from DESeq2and EdgeR are summarized in Additional file 5. Theconsistency between both methods was compared byPearson’s correlation coefficient (mean correlation be-tween methods 0.948 ± 0.01 SD) and Bland Altman

analysis [21] (mean difference between methods of 0.02and 97.37% of the measurements within the 95% confi-dence interval), which evidence high agreement betweenboth techniques (Additional file 6). We detected a dis-agreement between both methods in at least one BRCAsubtype in six genes (CRYAA, DNAJB13, DNAJC5G,HSPA6, HSPB3 and ODF1), all of which presented lowexpression levels. EdgeR runs with least computational re-sources than DESeq2, this motivated its preferential use.EdgeR ANOVA-like test was used to analyse differentialgene expression within PAM50 subtypes andHSP-Clusters (Additional file 7).

Heatmap construction and cluster analysisThe values of logarithm base 2 of normalized RSEM(RNAseq) plus 1 from 1033 patients (males, Normal-liketumours, and patients without clinical data were ex-cluded) from the TCGA cohort were used to constructthe HSPs expression matrix. The rows and columnswere sorted based on a hierarchical cluster with averagelinkage and Pearson’s correlation distance. According toSilhouette dendrograms analysis (Additional file 8) pa-tients were grouped into three clusters: HSP-Clust I,HSP-Clust II and HSP-Clust III.

Survival modelThe survivals analysis was performed according toREMARK guidelines [22]. The effect of each HSP onsurvival was estimated using a univariate Cox propor-tional hazard model with the survival information of the1033 patients of the TCGA cohort considered in theheatmap graphic and cluster analysis. To correct formultiple testing FDR testing was conducted byBenjamini and Hochberg method. Once each patient ofthe TCGA and the METABRIC training and test setwere classified into one of the three HSP clusters,Kaplan-Meier curves for each group were generated andthe survival distribution was compared using Log-Ranktest. A multivariate Cox proportional hazard model wasused to determine statistically significant survival differ-ence between clusters of TCGA cohort. The model wasadjusted to several known prognostic predictors (inclu-sion criteria): lymph node status, tumour size, age,tumour stage, and PAM50 subtypes. As exclusion cri-teria we considered: males, patients with unknown meta-static status at the time of diagnosis, and Normal-likesubtypes. From this filtering 1003 patients were left, with81 events registered. The sample size was not considereda priori and all available patient data within inclusioncriteria were considered.

Nearest centroid classifierTo train a HSP single-sample-predictor with the METABRICdataset, samples gene expression levels were scaled

Zoppino et al. BMC Cancer (2018) 18:700 Page 3 of 17

and only probes that were associated with the 95HSPs where used in the classifier. In cases wherethere was more than one probe matching a singlegene, all probes values were averaged and collapseinto one. From the 95 HSP genes, HSPA7 did notmatch to any of the probes analysed and those HSPgenes that presented low expression levels in theTCGA cohort (DNAJB8, DNAJC5G, DNAJB3, ODF1,CRYAA and HSPB3) were not considered to train theclassifier. The dataset was randomly divided into atraining set (n = 915) and a test set (n = 914), then ahierarchical clustering algorithm with average linkageand Pearson’s correlation distance was applied to thetraining dataset and the resulting dendrogram treewas cut to divide the set of patients into three differ-ent HSPs expression profile groups. From each clus-ter, the corresponding centroid vector was calculatedand the samples in the test set were labelled accord-ing to the class centroid from which each sample pre-sented highest Spearman correlation.

ResultsTranscriptomic analysis evaluating the RNA expressionprofile in TCGA BRCA cohortWe first evaluated the absolute normalized expressionlevels of the 95 HSP genes. The overall trend indicatesthat HSPs were highly expressed in tumour samples(one-sided Mann-Whitney U test P val = 1.256e− 10),nevertheless, a more detailed study showed a group ofsix genes (DNAJB8, DNAJC5G, DNAJB3, ODF1,CRYAA, and HSPB3) with very low expression levels inalmost all the samples and was not detected in at least50% of the cohort or more. On the other hand, six HSPs(HSP90AB1, HSP90AA1, HSPA8, HSP90B1, HSPA5, andHSPA1A) were ranked in the top 100 most expressedmRNAs of BRCA (hypergeometric test P val = 4.09− 07).(Fig. 1 and Additional file 1: Table S2). The rest of theHSPs were distributed in a wide range of expression. Al-most all members of the Chaperonin subfamily (TCP1,CCT2, CCT3, CCT4, CCT5, CCT6A, CCT7, and CCT8)were also expressed at similarly high levels. It is import-ant to note that the HSPB subfamily, except HSPB1, ap-peared with low transcript expression levels.We continued the analysis evaluating DEG comparing

BRCA tissues against normal tissues. In this study weconsidered only genes that showed absolute values oflog2 fold change (log2FC) > 1 and statistical significance(FDR < 0.05). The tabulated results (Additional file 5)show that in BRCA there were 3994 upregulated and2155 downregulated genes. To our knowledge, this isthe first report of the DEG between tumours and normaltissues taking into account PAM50 groups of RNAseqBRCA data (1097 patients). With respect to HSPgenes, 13 were upregulated and 11 were downregulated

(Additional file 1: Table S3). Deregulation of HSP genesincreased in BRCA subtypes as follows: Luminal A,Luminal B, HER2 and Basal. To achieve a better statisticalinterpretation volcano plots were used (Fig. 2). Thesegraphs allow the contextualization of the HSP genes re-spect to the rest of the genes letting a complete appreci-ation of gene expression changes that were modulateddifferentially in the entire cohort (Fig. 2 tumour total) andbetween the intrinsic BRCA subtypes (Fig. 2). The patientswere subdivided according to the PAM50 classification toinvestigate whether the intrinsic subtypes of BRCA mani-fested different expression of HSP genes. The PAM50classification is a “single sample predictor” and classifieseach of the samples in 5 tumour intrinsic subtypes [2].From a total of 1097 samples 566 were classified asLuminal A, 217 as Luminal B, 82 as HER2-enriched,192 corresponded to Basal and 40 were Normal-like(Additional file 1: Table S4). The comparison of thecorrelative immunohistochemical characteristics of eachtumour was included; these results appeared congruentwith the molecular classification (Additional file 1:Table S4). In the case of upregulated HSP genes, the log2fold-change mean and standard deviations (SD) in the dif-ferent subtypes ranged between 1.38 and 1.64 and 0.31 to0.69 respectively; the downregulated genes showed log2fold-change mean in the range of 2.34 to 3.62 and weremore dispersed (SD = 1.36 to 2.26) compared to upregu-lated genes. Surprisingly we found that several HSPs werewithin the first hundred genes with the lowest FDR valuesin Luminal A and Luminal B, which points out that someHSPs DEG in BRCA shows remarkable steady differencesbetween normal and tumour samples.After exploring HSPs expression changes, we found

many deregulated HSP genes, some of which were spe-cific for certain molecular subtypes while others wereshared by different intrinsic subtypes (Fig. 3). In particu-lar, this analysis revealed that 38 of the 95 HSP geneswere found differentially expressed. In the case ofdownregulated genes, a group (DNAJB4, DNAJC18,HSPA12A, HSPA12B, HSPB2, HSPB6 and HSPB7) pre-sented decreased transcript levels in all BRCA molecularsubtypes while some HSPs showed subtype specificdownregulation (DNAJC27 and DNAJC12 in Basal andBBS12 and DNAJC5G in HER2). Others HSPs presenteddecreased levels of transcripts shared between differentsubtypes (HSPB8 between HER2 and Basal and CRYABand SACS between HER2, Luminal A and Luminal B tu-mours). Evaluating the upregulated genes, we found amore complex combination where only DNAJC5B wasupregulated in all subtypes. HSPB1, DNAJB13, DNAJC1and DNAJC22 were upregulated in all except in theBasal subtype. The Basal subtype showed the highestnumber of specific upregulated genes (DNAJC2,DNAJC6, HSPA5, HSPA14 and CRYAA), DNAJA3 and

Zoppino et al. BMC Cancer (2018) 18:700 Page 4 of 17

CCT2 were upregulated in Luminal B, and DNAJB3 wasonly upregulated in HER2 tumours. Luminal A did nothave any specific upregulated HSP.

Fold change expression values of the different HSPsubfamilyWe next proceeded to compare the magnitude the HSPsDEG pattern in the BRCA tissues arranging the HSPs intheir five subfamilies. Figure 4 shows that the CHAPsubfamily (14 members) appeared upregulated in BRCAwith only three members (BBS10, BBS12 and in a lesserdegree CCT6B) downregulated. In this figure we canalso see that most of the HSP70 subfamily memberswere upregulated while only two members (HSPA12Aand HSPA12B) were strongly downregulated. HSPA4Lshowed a particular profile, its expression decreased inHER2 and Luminal A cancers only. The study of theHSPB subfamily showed interesting characteristics.Incremented transcripts levels of HSPB1, HSPB9 andHSPB11 were observed in most BRCA subtypes, CRYAA

was upregulated only in Basal subtype and ODF1showed an increased expression in Luminal A tumoursthat was not significant by the Deseq2 method. Interest-ingly, the genes CRYAB, HSPB2, HSPB6 and HSPB7were strongly downregulated in all BRCA subtypes. TheHSPC subfamily involves HSP90 genes with well-knownclinical implications in cancer [23]. HSPC membersshowed mild positive fold changes in all BRCA subtypes.It is of interest to mention that several HSP genes haverelatively high expression levels in normal tissues,therefore in these cases fold changes in expressionlevels between normal and cancer tissues are less pro-nounced but could be of important biological signifi-cance. (e.g. HSP90AA1 have a fold change of 0.98).The large DNAJ subfamily revealed a mixed behav-iour, some members (DNAJA2, DNAJB1, DNAJB8,DNAJB9, DNAJC8, DNAJC25) showed null variations,others were upregulated (DNAJA1, DNAJA3, DNAJA4,DNAJB2, DNAJB11, DNAJC1, DNAJC2, DNAJC5,DNAJC5B, DNAJC9, DNAJC10 and GAK) and some were

a b

Fig. 1 HSPs expression in breast cancer. a) The mean expression of each gene in all cancer samples was calculated and sorted in decreasing order.HSP genes were localized with a red x. Note that six HSP genes are above the orange line of the top 100 expressed genes. b) The graphs show theRNA expression distribution of HSP genes in the cohort. Note that figure is thicker were the values are more frequent

Zoppino et al. BMC Cancer (2018) 18:700 Page 5 of 17

downregulated (DNAJB4, DNAJC18, DNAJC27, DNAJC28and SACS) in all subtypes. Several interesting expressionprofiles of DNAJ members need to be especially mentioned.For example, DNAJC12 appeared strongly upregulated inLuminal A and B, in contrast to the Basal subtype wherethis gene appeared downregulated. DNAJB3 transcripts ap-peared strongly upregulated in the HER2 BRCA subtypeand DNAJC22 appeared upregulated in Luminal A,Luminal B and HER2 subtypes. A summary of the HSPsubfamilies fold change trends across PAM50 classes isdepicted in Additional file 9 to grasp a better understandingof the HSP groups changes and variability in the different

subtypes which reveals that a complex regulation isactive on every HSP subfamily, even for members ofthe same group.Beyond particular cases, less marked but important

differences were found in the overall expression patternsof HSP gene families between subtypes. Primarily, HSPH(from the HSP70 superfamily), HSP90 (HSPC), and typeI and type II chaperonins (from the CHAP family) werefound expressed at higher levels in Luminal B, HER2and Basal tumours than in Luminal A subtypes, whilefor the HSPB family, Basal tumours showed an overallless marked decrease of these group of genes with

Fig. 2 Differential expression of total genes in breast cancer. Volcano plots of genes expression analysis accomplished by Edge R method. In thex-axis the log2 fold change respect to normal tissue is represented, while in y-axis the -log10 of FDR is shown (the higher values show smaller FDR).Observe that HSP genes with log2 fold change > 1 and FDR < 0.05 are indicated as red circles. The green symbols at the top of the subpanels indicategenes with very small FDR (FDR < 5e− 324). Significant fold changes of non-HSP genes are light blue coloured

Zoppino et al. BMC Cancer (2018) 18:700 Page 6 of 17

respect to normal tissue, which represents greater ex-pression of them with respect to the rest of the subtypes,especially in relation to HER2 and Luminal B types(Additional file 10 A).

HSPs expression variability and clinical outcomeTo investigate whether the complex regulation of HSPgenes was associated with clinical outcome, we per-formed an integrated transcriptomic analysis of the 95HSP genes in the TCGA BRCA patients with knownfollow-up (n = 1033; Normal-like subtypes excluded). Itis well-known that several HSPs have clinical correlates,the best example is probably HSP90AA1 that it is usedas an adverse prognostic factor not only in BRCA butalso in other cancers [23]. In order to get further infor-mation of the clinical relevance of HSPs, we performedan overall survival analysis by Cox univariate modelbased on the expression levels of each HSP. We

observed 23 HSP genes with clinical statistical signifi-cance from which five genes were associated with a goodprognosis (HSPA2, DNAJB5, HSCB, HSPA12B andDNAJC4) and 18 (CCT6A, DNAJA2, HSPA14, CCT7,HSPD1, CCT2, HSPA4, DNAJC6, CCT5, SEC63, HSPH1,CCT8, CCT4, HSP90AA1, HSPA8, DNAJC13, HSPA9and TCP1) with a poor prognosis (Table 1).Next, we explored whether the BRCA patients could be

grouped into clinically relevant clusters based on HSPs ex-pression patterns. To test this hypothesis we performedan unsupervised hierarchical cluster analysis that sepa-rated the TCGA cohort into three main branches (Fig. 5).The three groups were called HSP-Clust I (red in Fig. 5),HSP-Clust II (green) and HSP-Clust III (orange). Thesethree HSP clusters corresponded to PAM50 classificationas follows: the HSP-Clust I had 83% of Luminal A tu-mours, HSP-Clust II was composed mainly by Basal-liketumours (92%), and the HSP-Clust III was the most het-erogeneous group with 44% of Luminal A tumours and40% of Luminal B tumours (Fig. 6a). The HER2 subtypewas dispersed into the three HSP groups, but the majoritywere seen in the HSP-Clust III. The Kaplan-Meier curvesof the HSP clusters showed highly significant differencesin overall survival between groups (Fig. 6b, P = 0.0022),letting us identify a low-risk group (HSP-Clust I) and ahigh-risk group (HSP-Clust III). Multivariable analyses ofHSP-Clust I against HSP-Clust II and HSP-Clust III ad-justed for known clinical covariates (tumour size, nodestatus, age, and tumour stage) showed different survivalrates for the HSP-Clust II, with a hazard ratio = 2.829 (CI95% = 1.55–5.17) and P value = 0.0007; and HSP-Clust IIIhazard ratio = 2.003 (CI 95% = 1.18–3.39) and P value =0.01 (Fig. 7a). We also tested a model including the intrin-sic molecular subtypes.In this case the P values ofHSP-Clust coefficients became non-significant (Fig. 7b),which suggests that HSP-Clusts effect on survival is re-lated to PAM50 subtypes. In order to validate theHSP-Clusts found, we used the METABRIC cohort di-vided in a training and test set to reproduce our results.Briefly, by a hierarchical cluster algorithm we divided thetraining set into three distinct groups which were consist-ent with the HSP-Clusts found in the TCGA dataset(Additional file 11 A) (TCGA HSP-Clust I vs. METABRICHSP-Clust I with a correlation factor = 0.87, TCGAHSP-Clust II vs. METABRIC HSP-Clust II with a correl-ation factor = 0.82 and TCGA HSP-Clust III vs. METAB-RIC HSP-Clust III with a correlation factor = 0.7).Centroids for each HSP-Clusts from the training set wereused to classify samples from the test set. The centroidsobtained from the test sets were in agreement with theothers centroids (Additional file 11 A). The PAM50 sub-type distribution regarding HSP-Clusts was similar in bothsets (Additional file 11 B). The overall survival of theHSP-Clusts corresponding to training and test sets showed

a

b

HER2

Luminal A Basal

Luminal B

CCT2DNAJA3

DNAJC12

DNAJA4HSPH1

DNAJC5BDNAJB13 DNAJC1DNAJC22 HSPB1

DNAJC2DNAJC6

HSPA5 HSPA14 CRYAAHSPA6

HYOU1 DNAJB11CCT6A

CCT5CCT3

HSPE1 DNAJC9 HSPD1

DNAJB3

HER2

Luminal A

Luminal B

Basal

BBS12

HSPB8

DNAJC5G

CRYABSACS DNAJB4

DNAJC18HSPA12AHSPA12BHSPB2HSPB6HSPB7

DNAJC12DNAJC27

up

-reg

ula

ted

do

wn

-reg

ula

ted

Fig. 3 Venn diagrams showing overlapped and specific differentiallyexpressed HSPs in intrinsic subtypes of breast cancer. The figureshows a summary of HSP genes expression analysis performed byEdge R method (fold-change > 2, FDR-adjusted P values < 0.05,and with no disagreement mean between the EdgeR andDESeq2 methods). Normal group was discarded based on thelow number of cases. a Down-regulated HSP genes. b Up-regulated HSP genes

Zoppino et al. BMC Cancer (2018) 18:700 Page 7 of 17

a significant difference between HSP groups (both trainingand test set had a Log-Rank test with a P value < 0.0001)(Additional file 11 C).It is interesting to note that there is a significant (but

not complete) overlap between BRCA PAM50 intrinsicsubtypes and HSP-Clusts. For instance, HSP-Clust I isenriched with Luminal A tumours and also presentslower expression levels of HSPH, HSPC and type I andII chaperonins compared to HSP-Clust II and HSP-ClustIII, which are enriched with Basal and Luminal B tu-mours respectively. HSP-Clust II presents significantlyhigher levels of some HSPB genes such as HSPB2,

HSPB3, CRYAA and CRYAB compared to the othersHSP subtypes (a pattern that was also observed inBasal-like tumours). HSP-Clust III is enriched withDNAJA gene expression (similar to the Luminal B andHER2 subtypes) (Additional file 10 B).

DiscussionThis is the first comprehensive study examining thewhole HSP family in breast cancer patients. The HSPfamily, characterized by 95 genes and one pseudogene,represents only 0.46% of the 20,531 analysed genes. Inthis study, we found that in BRCA almost 30% of the

Fig. 4 Diagram showing a summary of HSPs expression grouped in subfamilies in breast cancer according to the intrinsic molecular subtypes.In the figure, the diameter of the circles shows the log2 fold change assessed by EdgeR method. The circles in green show downregulated genesand the red ones represent upregulated genes. The circle opacity is related to the FDR values, circles with FDR > 0.05 are transparent and thereforenot depicted. The figure makes emphasis on fold change expression values regardless any threshold

Zoppino et al. BMC Cancer (2018) 18:700 Page 8 of 17

total genes were deregulated (19.45% upregulated and10.5% downregulated), where the HSP family accountsfor 0.39% of this deregulation (0.32% of the upregulatedgenes and 0.52% of the downregulated). Several reasonshave been mentioned to explain HSP misregulation incancer: by the stressful situations found in cancer tissues[4], to increase the stabilization of transcription factors,receptors, protein kinases and other proteins that liealong the pathways of normal to cancer transition [24],and by the oncogenic agents/events that directly affectthe heat shock response [25]. The activation of HeatShock factors (HSF) during cancer progression can inturn explain the activation of the HSPs molecular chap-erones [26, 27]. Therefore, considering that cancer tis-sues are subjected to several stressful situations weexpected to see more upregulated HSPs (n = 13) andfewer downregulated (n = 11). At this point we have tosay that the expression levels of several HSPs were veryclose to the cut-point used (log2 fold-change = ±1), thishappened for example with the HSPC family whichcodes for the HSP90 (all appeared with a certain level of

upregulation, see Fig. 3). In any case, it is evident that inBRCA the expression levels of several HSP family mem-bers are affected. Upregulation was noted mainly in theCHAP and HSPC family members while the greatestdownregulation was observed in most HSPB members(Fig. 3 and Additional file 9). The downregulation of thesmall HSPs agrees with a recent report [28]. The HSP70superfamily (which includes the HSP70 and HSP110 orHSPH family) and the DNAJ members showed variableresults with ups and downs.The present study revealed that deregulation of the

HSPs varied according to the BRCA molecular subtype.Of importance at this point is: what are the functionalimplications of the up- and down-regulation of the HSPgenes in each breast cancer subtypes? This is not an easypoint to address because in the present report we arefinding alterations in HSP genes that are little known tobe linked with breast cancer; moreover others likeDNAJB3 (increased in HER2 subtype), DNAJB13 andDNAJC22 (increased in Luminal and Basal subtypes),and SACS (increased in all subtypes) have not beenrelated with any cancer type. Let’s begin with theChaperonin family. The members of this group can be di-vided into three distinct subgroups: Type I chaperonins,established by HSPE1 and HSPD1 genes (also known bytheir bacterial names GroES and GroEL or HSP10 andHSP60 respectively), type II chaperonins forming theT-complex protein-1 ring complex (TRiC) which isformed by a double ring structure with eight distinctsubunits (TCP1 and CCT genes) working as an ATPdependent protein folding machinery [29], and finally theBBS group of genes (BBS10, BBS12 and MKKS) that inconjunction with the TRiC complex mediate the BBSomeassembly [30]. Of this group of genes, HSPD1, HSPE1,CCT3 and CCT5 were overexpressed in Basal, HER2 andLuminal B subtypes (more aggressive BRCA tumours).HSPD1 and HSPE1 are located on chromosome 2arranged in a head-to-head orientation and both are im-plicated in macromolecular protein assembly and mito-chondrial protein import, while CCT3 and CCT5 form aprotein complex folding various proteins including actinand tubulin upon ATP hydrolysis and, as part of the BBS/CCT complex, they are involved in the assembly of theBBSome, which in turn is implicated in ciliogenesis regu-lating transports vesicles to the cilia [30]. At this point wehave to remember that breast cancer cells, mainly stemcells, have primary cilia (a non-motile microtubule basedcell-surface organelle) that acts as a cellular antenna forreceiving signaling pathways involved in the regulation ofcell proliferation, differentiation and migration [31, 32].Therefore our study adds evidence to an important role ofCCT3 and CCT5 in the more aggressive BRCA tumours:Basal, HER2 and Luminal B subtypes. CCT3 has been in-volved in mitosis progression and associated with poor

Table 1 Univariate Cox proportional hazard risk of breast cancerbased on HSP expression. Regression coefficients, hazard riskcoefficients, standard error, P value and FDR are presented.Only HSP genes with FDR < 0.05 are shown

Gene Coefficient HR Coeff SE P-val FDR

HSPA2 −0.35 0.71 0.10 < 0.001 0.005

DNAJB5 −0.32 0.73 0.10 0.002 0.011

HSCB −0.29 0.75 0.11 0.009 0.037

HSPA12B −0.29 0.75 0.10 0.003 0.016

DNAJC4 −0.27 0.76 0.10 0.006 0.027

CCT6A 0.22 1.25 0.08 0.009 0.037

DNAJA2 0.25 1.29 0.08 0.002 0.011

HSPA14 0.27 1.32 0.09 0.001 0.009

CCT7 0.28 1.32 0.11 0.008 0.034

HSPD1 0.30 1.35 0.10 0.003 0.013

CCT2 0.30 1.35 0.08 < 0.001 0.001

HSPA4 0.31 1.36 0.11 0.005 0.025

DNAJC6 0.34 1.40 0.11 0.002 0.011

CCT5 0.35 1.42 0.10 < 0.001 0.005

SEC63 0.35 1.42 0.09 < 0.001 < 0.001

HSPH1 0.35 1.42 0.10 < 0.001 0.004

CCT8 0.40 1.49 0.10 < 0.001 0.001

CCT4 0.40 1.49 0.10 < 0.001 < 0.001

HSP90AA1 0.40 1.49 0.09 < 0.001 < 0.001

HSPA8 0.41 1.51 0.12 < 0.001 0.004

DNAJC13 0.46 1.58 0.11 < 0.001 < 0.001

HSPA9 0.46 1.58 0.10 < 0.001 < 0.001

TCP1 0.50 1.64 0.10 < 0.001 < 0.001

Zoppino et al. BMC Cancer (2018) 18:700 Page 9 of 17

prognosis in hepatocellular carcinoma [33], has been im-plicated in osteosarcoma tumorigenesis [34], and appearedas a candidate biomarker in epithelial ovarian cancer [35]and in cholangiocarcinoma patients [36]. CCT3 was founddifferentially expressed in colon and other epithelial can-cers [37] and its expression has been associated with drugresistance in a squamous lung cancer cell line [38]. CCT5was found upregulated in p53-mutated breast tumoursand might be implicated in resistance to docetaxel treat-ment [39]. Of notice, all the other TRiC genes except

CCT6B were also among the most highly expressed incancer and upregulated accordingly in the different sub-types, suggesting an important role of the TRiC complexspecifically in BRCA as previously suggested [40]. TRiChas an essential role in cell proteostasis in physiologicalconditions but also in oncogenesis and cancer progression[41] and is known to regulate the proper folding of severalothers genes involved in cancer such as actin, tubulin [42],p53 [43] and protoncogene STAT3 [44]. In our study,HSP-Clust II (enriched with Basal-like tumours) presented

CHAPDNAJHSPA HSPB HSPC

Subfamilies: BadGood

Prognosisassociation

FDR < 0.05−5 0 5

0

4000

Co

un

t

Scaled expression

PR

THER2

ER

PAM50

HSP-CLUSTER

BasalLumA

Her2LumB N0 N+N

PositiveNegativeIndeterminated

NA

<2 >2

HSP-Clust I HSP-Clust II HSP-Clust III

HSPA6HSPA7

DNAJB13HSPB9

DNAJB12HSPB1

GAKHSCB

DNAJB2DNAJC17

DNAJC4DNAJC30DNAJC15

DNAJB5CRYABHSPB2

HSPA12BHSPB6HSPB7

DNAJC5BDNAJC7HSPB11DNAJC8

MKKSDNAJC5

DNAJC11DNAJA3

TRAP1DNAJB1HSPA1AHSPA1BDNAJB6

SEC63DNAJC21

HSPA4HSPA9

DNAJA2DNAJB11

HYOU1HSPA5

HSP90B1DNAJC9DNAJC2HSPA14

HSPA8DNAJA1HSPH1

HSP90AA1CCT2CCT3

HSP90AB1CCT8TCP1CCT5

CCT6ACCT4CCT7

HSPD1HSPE1

DNAJB7DNAJB4

SACSDNAJC6HSPA4L

BBS12DNAJC18DNAJC24

DNAJB9DNAJC3

DNAJC16DNAJC27DNAJC10

HSPA13BBS10

DNAJB14DNAJC13DNAJC22HSPA12A

HSPB8DNAJA4CCT6B

DNAJC28HSPA2

DNAJC25HSPA1L

DNAJC12DNAJC14

DNAJC1DNAJC19

Fig. 5 HSPs gene expression heatmap of TCGA BRCA cohorts. Expression patterns of 89 HSP genes in 1033 samples are depicted (central panel,low expression levels in blue and high expression levels in red). By a hierarchical clustering algorithm patients were group into HSP-Clust I (red),HSP-Clust II (green) and HSP-Clust III (orange) (upper dendrogram). Several rows were added to indicate: immunohistochemical status of receptors (ER,PR and HER2), tumour size (T > 2 cm or T < 2 cm), satellite nodules spread (N positive or N negative) and PAM50 classification. We also added threecolumns indicating HSP corresponding subfamilies, univariate Cox’s regression model coefficients (pink represents positives coefficients(bad prognosis), while light blue are negatives coefficients (good prognosis)) and its corresponding FDR values (black boxes representFDR value for Cox’s coefficients < 0.05)

Zoppino et al. BMC Cancer (2018) 18:700 Page 10 of 17

high expression levels of the TRiC complex genes. Thecurrent standard of treatment of triple-negative (TNBC) tu-mours is systemic neoadjuvant chemotherapy that typicallyinclude taxanes which inhibit tubulin depolymerization[45]. We hypothesize that the measurement of the TRiCcomplex genes along with the classification of tumour sam-ples in the different HSP-Clusts could be used as an im-portant tool to predict taxane response, even thoughfurther studies are needed to validate this assumption.Coming back to HSPE1, in a previous proteomic ana-

lysis this protein appeared with altered expression inMDA-MB-231 breast cancer cells (triple negative highlyaggressive cells) [46] and both HSPD1/HSPE1 have alsobeen found upregulated in other cancer types associatedwith tumour cell transformation [47]. Interestingly, bothTRiC genes and HSPD1/HSPE1 were co-expressed andwere associated with worst prognosis individually andhad high expression in the HSP-Clust II and III of our

study (Additional file 10 B). All this data together sug-gest that not only the TRiC complex has a protagonistrole in cancer behaviour but also that the HSPD1/HSPE1 complex is involved tightly with TRiC in proteos-tasis regulation, an association that is poorly understoodin breast cancer and should be further studied. On theother hand, BBS12 was underexpressed in the HER2subtype predominantly and along with BBS10, bothshowed decreased expression levels in all subtypes. MKKSgene (also known as BBS6) was not altered. Therefore, ourstudy reveals specific chaperones that participate in the as-sembly of the BBSome altered in BRCA.The HSP70 family is a group of evolutionary con-

served and ubiquitously expressed genes that in con-junction with the DNAJ family act as a protein foldingregulatory network that also protects the cell againststressful conditions [48]. Several members of the HSP70family were found highly expressed (HSPA8, HSPA5,HSPA1A) or upregulated in BRCA. We found thatHSPA6 expression appeared elevated mainly in LuminalA, Luminal B and Basal subtypes. In a previous studyhigh levels of this protein were associated with recur-rence in hepatocellular carcinoma [49]. HYOU1 alsoknown as oxygen-regulated protein 150 (ORP150) wasupregulated in HER2 and Basal subtypes and the proteinhas been implicated with tumour progression in differ-ent cancers [50–53]. HSPA5 was found highly expressedin all subtypes, and especially upregulated in Basal tu-mours in our study, and has been associated with endo-plasmic reticulum stress response (ERSR), inhibition ofapoptosis and autophagy in several studies [54–56].HSPA8 was the most expressed gene of the HSP70 fam-ily and one of the genes with the strongest associationwith survival in our study. This gene is constitutivelyexpressed and has been largely associated with theprotein folding and stress response [57, 58]. Interest-ingly, DNAJC12, a gene strongly upregulated inLuminal A and B tumours, was found to interact withHSPA8 under ERSR [59].Only one HSP appeared upregulated in the four sub-

types considered: the protein encoded by DNAJC5B,which is implicated in protein processing at the level ofthe endoplasmic reticulum [60]. This protein has beenfound in secretory vesicles as well as in synaptic andclathrin-coated vesicles in neuroendocrine, exocrine andnervous cells. Of interest is that this member of theDNAJ family has been found upregulated in humanbladder carcinoma, gastric adenocarcinoma, and glio-blastoma cell lines by the OCT4B1 variant (octamer-binding transcription factor 4 B1 variant) which isexpressed by pluripotent normal and cancer stem celllines and linked to anti-apoptosis [61]. In addition, theseauthors found that the OCT4B1 variant is also linked toupregulation of the chaperonin DNAJC11 which is

a

b

Lum A(n=553)

Basal(n=191)

HER2(n=80)

LumA382HSP-Clust I

LumB

2HSP-Clust II

Basal

169HSP-Clust III

Her2525

152

6178

7

199

52

Lum B(n=209)

Clu

st I

Clu

st II

Clu

st II

I

p = 0.00220

25

50

75

100

0 500 1000 1500 2000 2500 3000 3500time (days)

surv

ival

(%

)

+

HSP-Clust I

HSP-Clust II

HSP-Clust III

459 304 142 103 69 53 33 14

235 104 78 57 33 23 16194 121 53 42 32 21 16 6380

0 500 1000 1500 2000 2500 3000 3500time (days)

Number at risk

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++ +++++++++++++++++++ +

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++ ++++++++++ ++++++++++++

+

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++ ++ +++++++++++++++++++++ ++++++ +++

Fig. 6 HSP cluster characterization. a) Agreement between PAM50and HSP clusters. The size of the bars is in proportion to the numberof samples in each category. b) Overall survival of HSP clusters. Kaplan-Meier curves corresponding to HSP-Clust I, HSP-Clust II and HSP-ClustIII. Statistical significance was evaluated by Log-Rank test

Zoppino et al. BMC Cancer (2018) 18:700 Page 11 of 17

complexed with mitofilin in the mitochondrial mem-brane [62] and has been associated with neuromusculardiseases and lymphoid abnormalities [63]. In this study,DNAJC11 appeared slightly upregulated in Luminal B,HER2 and Basal subtypes. No attention has beendirected to these proteins (DNAJC5B and DNAJC11) inBRCA. It is now evident that further studies must be di-rected to clarify the role of these proteins. DNAJC9 ap-peared upregulated in Basal, HER2 and Luminal B, andin previous studies has been found upregulated innode-positive uterine cervical carcinoma [64].Our study revealed HSPs that appeared both deregu-

lated and not well studied in BRCA; for example,DNAJB3 appeared with high levels of upregulation onlyin HER2 BRCA subtype. Close gene location with HER2gene cannot explain upregulation of DNAJB3 since thisgene is located on chromosome 2 while HER2 (amplifiedin HER2 subtype) is located on chromosome 17. Little isknown about the protein encoded by this gene, and itsrole in cancer in general and in breast cancer in particu-lar is not known. DNAJB3 has been reported downregu-lated in obese human subjects, DNAJB3 over-expressionin adipose cell lines caused: a) reduction in JNK (JunN-terminal kinase) improving insulin sensitivity and en-hancing glucose uptake and b) mediated PI3K/AKTpathway activation [65]. Of interest here is that thePI3K/Akt signalling pathway is negatively regulated byPTEN and we have reported that PTEN is downregulated

by HSPB1 (HSP27), both proteins have been implicated inHER2-positive tumours [66]. Therefore, it will be of inter-est to study the role of DNAJB3 in HER2 BRCA. However,we have to take into account that the upregulation levelsof this gene might appear statistically significant, but thenumber of RNA molecules could be relatively low. There-fore, an upregulated gene could have few RNA copy num-bers and we ignore if the encoded protein has biologicalsignificance. Nevertheless, this entire complex HSP70/DNAJ landscape suggests an intricate regulatory inter-action between these genes that remains to be untangled.Finally, among the upregulated small heat shock pro-

teins, HSPB1 stands out as the highest expressed of thegroup and appeared upregulated in Luminal A, LuminalB, and HER2 (close to the cut-point in Basal); the pro-tein encoded by this gene has been well studied in breastcancer [4, 67].Many of these upregulated genes and proteins

have been reported as associated with tumour pro-gression in different cancer types and in several op-portunities with poor prognosis. In concordance, wehave found that some of these genes appeared up-regulated mainly in aggressive breast cancer sub-types that were clustered in the HSP-Clust IIIgroup. Moreover, the complexity of the regulation ofthe HSPs in BRCA is further increased when we considerthe high number of client proteins that are associatedwith the HSPs [11].

Variables

HSP-Clust II vs HSP-Clust IHSP-Clust III vs HSP-Clust IAgePositive node statusTumor size >2 cmTumor stage IITumor stage IIITumor stage IVLumB vs LumAHer2 vs LumABasal-like vs LumA

No of Patients (%)

190 (18.94%)361 (35.99%)58.28(mean)

483 (48.16%)744 (74.18%)582 (58.03%)232 (23.13%)

17 (1.69%)203 (20.24%)

75 (7.48%)188 (18.74%)

Hazard Ratio (95% CI)

2.508 (0.68−9.24)1.608 (0.9−2.88)

1.042 (1.02−1.06)0.577 (0.3−1.13)0.974 (0.45−2.1)

1.285 (0.41−4.02)2.228 (0.57−8.71)

3.827 (0.81−18.02)1.338 (0.71−2.53)3.227 (1.53−6.8)

1.239 (0.33−4.61)

p−value

0.1670.111

5.22e−060.1080.9460.6670.249

0.08960.369

0.002070.749

1 2 3 4 5

Hazard Ratio

b

Variables

HSP-Clust II vs HSP-Clust IHSP-Clust III vs HSP-Clust IAgePositive node statusTumor size >2 cmTumor stage IITumor stage IIITumor stage IV

No of Patients (%)

190 (18.94%)361 (35.99%)58.28(mean)

483 (48.16%)744 (74.18%)582 (58.03%)232 (23.13%)

17 (1.69%)

Hazard Ratio (95% CI)

2.829 (1.55−5.17)2.003 (1.18−3.39)1.038 (1.02−1.06)0.601 (0.31−1.17)0.92 (0.43−1.98)

1.515 (0.49−4.68)2.588 (0.67−10.01)4.757 (1.03−21.99)

p−value

0.0007270.00964

2.45e−050.1320.83

0.4710.168

0.0459

1 2 3 4 5

Hazard Ratio

a

Fig. 7 Multivariable Cox Model of HSP-Clusts. a) Forest plot showing the hazard risk of HSP-Clusters controlling for confounders (age, node status,tumour size, tumour stage). Hazard ratios, 95% confidence interval and corresponding P values are depicted. b) Same Cox’s model plus de addition ofPAM50 subtypes as covariates

Zoppino et al. BMC Cancer (2018) 18:700 Page 12 of 17

Another interesting observation from the presentstudy is that several HSPs were downregulated in allbreast cancer subtypes: DNAJB4, DNAJC18, HSPA12A,HSPA12B, HSPB2, HSPB6, and HSPB7. DNAJB4 is amember of the DNAJ family and is described as atumour suppressor [68], which is in agreement with ourresults; increased expression of DNAJB4 has been impli-cated in the stabilization of wild-type E-cadherin (butnot the mutant) stimulating the anti-invasive function ofE-cadherin in gastric cancer cells [68]. Little is knownabout the protein coded by DNAJC18, but a poly-morphic variant has been associated with aggressivebladder carcinoma [69]. HSPA12A encodes a protein ofthe HSP70 family that seems to act like a protectivefactor in gastric cancer [70]. We found high levels ofsuppression in several members of the HSPB family(CRYAB, HSPB2, HSPB6 and HSPB7) (Fig. 3); in an inte-grated genomic and epigenomic analysis the ATM,HSPB2 and CRYAB (this last downregulated in LuminalA, Luminal B and Basal) genes were found commonlydeleted and underexpressed in patients with breast can-cer brain metastasis [71]. The role of CRYAB gene(Alpha B-crystallin HSPB5) is controversial in cancer[72–79], its expression has been associated with aggres-sive breast cancer subtypes. In agreement with ourresults, HSPB6 and HSPB7 have been found downregu-lated in several tumour types [80–85], and we reporthere this downregulation in all subtypes of BRCA is pos-sibly supporting a role as tumour suppressor genes. Inour analyses we compared tumour tissue with normalbreast tissue, but displacement of stroma in the tumoursamples could be affecting the results. Nevertheless, in arecent publication none of the HSP genes were found al-tered by the confounding effect of tumour purity [86].The HSPs expression patterns of the molecular subtypesare still heterogeneous [15] and the results of the presentstudy contribute to the characterization of these sub-types. We are now completing the study of the methyla-tion status of the HSP genes as well as the mutations,amplifications and deletions in these genes.Of importance, we have to mention that some genes

evaluated in this work presented clinically and biologic-ally meaningful characteristics already described, butsome others genes are totally unknown at the moment[87]. The clinically important genes DNAJB5, HSCB,HSPA2 (usually differentially overexpressed in LuminalA and B), DNAJC4, and HSPA12B (downregulated inBRCA) presented a significant FDR value in the Cox’sproportional hazard model presenting negative coeffi-cients (their expression was associated with a good prog-nosis). In contrast, the genes with high expression levelssignificantly associated with poor prognosis were:CCT6A, HSPA14, DNAJC6 (upregulated in Basal),CCT2 (upregulated in Luminal B), CCT5, HSPD1

(upregulated in Basal, Luminal B and HER2), SEC63(upregulated in HER2), TCP1, CCT4, CCT7, CCT8(upregulated in HER2 and Basal), HSP90AA1 (upreg-ulated with a near 0.9 log2 fold-change in Luminal B,HER2 and Basal), HSPH1 (upregulated in Luminal B,HER2), DNAJA2, HSPA9, HSPA4, DNAJC13, andHSPA8. Many of which were previously mentioned(HSP90AA1, TRiC, HSPD1/HSPE1, HSP70 family)and others for which their role in BRCA has not beenexhaustively studied.An important point of this study is the finding of three

discrete HSPs expression profiles with prognostic signifi-cance (P = 0.0022) that we called HSP-Clust I, II and III.These HSP clusters groups were reproduced in an inde-pendent dataset using the METABRIC cohort and a sin-gle sample predictor was trained to classify unknownsamples into one of the three HSP-Clusts with robust re-sults. Importantly, TCGA and METABRIC datasets weredeveloped using different RNA measurement technolo-gies but the clusters found showed striking similaritiesand had significant impact on disease outcome. Aninteresting point to address is that the HSP-Clust II(predominantly basal-like) in METABRIC is much moreclearly associated with a poor prognosis than the samesignatures in the TCGA, a plausible explanation mightbe found in the survival differences of Basal-liketumours in each cohort (Additional file 12). Even thoughHSP-Clusts survival is highly related to PAM50 subtypesas expected, it is important to notice that the overlap be-tween groups is not complete. Regarding Luminaltumours, HSP-Clust I presented mainly Luminal A tu-mours while HSP-Clust III presented mixed proportionsof Luminal A and Luminal B subtypes. These findingscould be reflecting differences in the biology of LuminalA tumours from HSP-Clust I with respect to Luminal Atumours of HSP-Clust III. Also, since HSPs have beenlong related with drug resistance, it would be of interestto test if the different HSP-Clust are related with differ-ent chemotherapy response profiles, which in turn,could imply a differential treatment for each HSP-Clustgroup. Further studies will be necessary to turn this clas-sification useful for clinical practice and to bettercharacterize the prognostic and treatment for thesegroups of patients. Since we used a combination of allHSP genes to evaluate survival, this could add superflu-ous information that can reduce the performance of thestudy. It will be interesting to reduce the number ofHSP genes in order to increase the potential of the HSPsexpression patterns as a prognostic factor. For in-stance, the clinical subset of HSP genes with clinicalimportance could be used as a genetic signature todevelop prognostic tests or as a base for future re-search of predictive assays based on immunohisto-chemistry, microarray or rPCR.

Zoppino et al. BMC Cancer (2018) 18:700 Page 13 of 17

ConclusionsOur results show the existence of several HSP genesderegulated in all molecular subtypes of breast cancerwhile others appeared deregulated in specific molecularsubtypes. We also found that the overall survival ofbreast cancer patients appeared associated with the ex-pression level of certain HSPs.

Additional files

Additional file 1: Table S1. Clinical data of TCGA patients. The datawas updated with the available follow up information (May, 2015).Table S2. Gene mean expression in breast cancer tissues. The meanexpression of each gene in all cancer samples was calculated andsorted in decreasing order. Table S3. Summary of misregulated genes inBRCA. Tabulated data show the number and percentages of total genesand HSP genes presenting > 2 fold-change in total samples and accordingto intrinsic BRCA subtypes. Table S4. Summary of PAM50 classification andimmunohistochemical characteristics of tumours. (XLSX 1080 kb)

Additional file 2: Data analysis workflow. Schematic representation ofHSPs transcriptomic and survival analysis process. (PDF 108 kb)

Additional file 3: PAM50 classification quality control of TCGA’s samplesI. A) Principal components analysis of the training and test sets. Note thesubtype clustering and the superposition between both datasets. B)Correlations between subtype assigned and the corresponding subtypecentroids per sample and relation between subtypes and proliferationindex. Each dot represents a single sample. (PDF 166 kb)

Additional file 4: PAM50 classification quality control of TCGA’s samplesII. Unsupervised hierarchical clustering of samples according to PAM50gene set expression. Note the consistency between the subtype assignedto each sample by PAM50 algorithm and the group composition determinedby the clustering technique. (PDF 133 kb)

Additional file 5: Differential gene expression of 20,531 genes comparingcancer tissue against normal breast tissue. The values were determined byEdgeR and DESeq2 methods; also the analysis was performed accordingmolecular subtype classification. For better data exploration HSP genes wereseparated in auxiliary tables. (XLSX 22435 kb)

Additional file 6: Fold-change consistency between EdgeR and DESeq2methods. A) Correlation analysis between fold-change obtained by bothmethods. The figure shows a tight linear trend between EdgeR and DESeq2fold-change estimations. Genes found significant for both methods arerepresented in yellow circles, in green and red are genes significantlydifferentially expressed by one of the two methods and in white, genes withno significant changes by both techniques. B) Bland Altman analysis comparingthe mean fold-changes of both methods (x-axis) and the difference betweenthem (y-axis). This plot allows the identification of any systematic differencebetween methods and possible outliers. Each circle represents an HSP geneand their colours the subtype for which the fold-change was calculated. Theblue dotted line represents the mean difference between both techniques(0.02) and the light blue dashed line depicts the upper (0.88) and lower (− 0.84)limits of the 95% confidence interval of the differences. (PDF 175 kb)

Additional file 7: HSPs differential gene expression between tumour tissues.The values were determined by EdgeR ANOVA-like method performed on20,531 genes from BRCA TCGA. Only HSPs values are showed. Each columnincludes log2 fold change values for all comparison, log2 mean counts permillion (logCPM), F-statistic and corresponding p-values and FDR values. Theconditions compared are Luminal A vs. Luminal B, Luminal A vs. HER2,Luminal A vs. Basal-like, Luminal B vs. HER2, Luminal B vs. Basal-likeand HER2 vs. Basal-like. Comparison between HSP-Clusts were alsoconsidered, namely HSP-Clust I vs. HSP-Clust II, HSP-Clust I vs HSP-Clust IIIand HSP-Clust II vs. HSP-Clust III. (XLSX 35 kb)

Additional file 8: Dendrogram analysis of hierarchical clustering basedon HSPs gene expression. The separation distance between branches wasdetermined by silhouette technique. The highest coefficient correspondsto the optimal number of cluster, in this case k = 3. (PDF 99 kb)

Additional file 9: Summary of HSP subfamily Fold Change trends acrossPAM50 subtypes. Boxplot representing HSP subfamilies log2 fold changeranges by EdgeR method in the different molecular subtypes of breastcancer. (PDF 111 kb)

Additional file 10: Differential gene expression in BRCA TCGA tumours.Summary of EdgeR ANOVA-like differential gene expression showing theHSPs pairwise differences between tumour subtypes. Genes were groupedaccording to their corresponding families. Chaperonins were divided intothree different types (type I, type II and BBs chaperonins), HSPH weredistinguished from the rest of the HSP70 family and DNAJ were dividedinto their three subfamilies (A, B and C). The vertical blue lines representsbaseline level from the reference subtype while the light blue points showsthe fold change of the HSP genes in each pairwise comparison. Red dotsare depicted for genes that had absolute log2 fold changes greater than 2.A) Shows the comparison between PAM50 molecular subtypes, and B)shows differences between HSP-Clust subtypes. (PDF 202 kb)

Additional file 11: HSP clusters characterization. A) Centroid of HSPclusters expression profiles for TCGA, METABRIC training and test set. Thecolour of the boxes in regard to the central dashed line represents down(blue) or upregulation (red) of the gene in the corresponding cluster. Thecontinuous black line represents the mean expression values of eachgene in the cluster compared to the mean of the same gene over allsamples. B) Agreement between PAM50 and HSP clusters for METABRICtraining and test sets. The size of the bars is in proportion to the numberof samples in each category. C) Overall survival of HSP clusters for METABRICtraining and test sets. Kaplan-Meier curves corresponding to HSP-Clust I,HSP-Clust II and HSP-Clust III. Statistical significance was evaluated byLog-Rank test. (PDF 214 kb)

Additional file 12: PAM50 subtypes overall survival in TCGA andMETABRIC cohorts. (PDF 166 kb)

AbbreviationsBRCA: Breast cancer; DEG: Differential expressed gene; ER: Oestrogen receptor;ERSR: Endoplasmic reticulum stress response; FDR: False discovery rate;HER2: HER2-enriched; HSP: Heat shock proteins; METABRIC: MolecularTaxonomy of Breast Cancer International Consortium; PR: Progesteronereceptor; TCGA: The Cancer Genome Atlas

AcknowledgmentsThe authors thank Dr. Katherine Hoadley (University of North Carolina ChapelHill) and Dr. Gary M. Clark (Boulder, CO) for advice. Also, gratefully acknowledgeDaniel Roden and Geoff Macintyre for their valuable comments and suggestionsthat improved the quality of the paper.

FundingThis work was partially supported by the following grants: Agencia Nacionalde Promoción Científica y Tecnológica PICT 2015 2607; CONICET PIP11220110100836 DAS 30844; Universidad Nacional de Cuyo SECTYP J062;and Universidad del Aconcagua (UDA). The authors confirm that the foundershad no influence over the study design, content of the article, or selection ofthis journal.

Availability of data and materialsThe TCGA datasets analysed during the current study are available in theGenomic Data Commons Data Portal (National Cancer Institute, NIH, USA)repository, (https://portal.gdc.cancer.gov/projects/TCGA-BRCA). The METABRICdata is available in the Synapse open source software platform under accessionnumber syn1757063, syn1757053 and syn1757055 (http://www.synapse.org).

Authors’ contributionsFCMZ and MEG have contributed equally to this work. Conception anddesign: FCMZ, MEG and DRC. Development of methodology: FCMZ andMEG. Acquisition of data: MEG. Analysis and interpretation of data:FCMZ, MEG, GNC and DRC. Writing, review and/or revision of themanuscript: FCMZ, MEG, GNC and DRC. Administrative, technical, andmaterial support: FCMZ, MEG, GNC, and DRC. All authors read andapproved the final manuscript.

Zoppino et al. BMC Cancer (2018) 18:700 Page 14 of 17

Ethics approval and consent to participateThis study has been approved by the Bioethical Committee of the MedicalSchool of the National University of Cuyo, Mendoza, Argentina (0029963/2015). The experiments comply with the current laws of Argentina in whichthey were performed.The results here are in whole or part based upon data generated by theTCGA Research Network (http://cancergenome.nih.gov/) and METABRIC(Molecular Taxonomy of Breast Cancer International Consortium). TheMETABRIC and TCGA offer anonymous data.

Consent for publicationNot applicable.

Competing interestsThe authors declare that they have no competing interests.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Received: 10 November 2017 Accepted: 20 June 2018

References1. Ferlay J SI, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM,

Forman D, Bray, F.: GLOBOCAN 2012, Cancer Incidence and MortalityWorldwide: IARC CancerBase No. 11; 2016.

2. Parker JS, Mullins M, Cheang MC, Leung S, Voduc D, Vickery T, Davies S,Fauron C, He X, Hu Z, et al. Supervised risk predictor of breast cancer basedon intrinsic subtypes. J Clin Oncol. 2009;27(8):1160–7.

3. Cancer Genome Atlas N. Comprehensive molecular portraits of human breasttumours. Nature. 2012;490(7418):61–70.

4. Ciocca DR, Calderwood SK. Heat shock proteins in cancer: diagnostic,prognostic, predictive, and treatment implications. Cell Stress Chaperones.2005;10(2):86–103.

5. Bozaykut P, Ozer NK, Karademir B. Regulation of protein turnover by heatshock proteins. Free Radic Biol Med. 2014;77:195–209.

6. Kennedy D, Jager R, Mosser DD, Samali A. Regulation of apoptosis by heatshock proteins. IUBMB Life. 2014;66(5):327–38.

7. Kampinga HH, Hageman J, Vos MJ, Kubota H, Tanguay RM, Bruford EA,Cheetham ME, Chen B, Hightower LE. Guidelines for the nomenclature ofthe human heat shock proteins. Cell Stress Chaperones. 2009;14(1):105–11.

8. Ciocca DR, Fanelli MA, Cuello-Carrion FD, Castro GN. Heat shock proteins inprostate cancer: from tumorigenesis to the clinic. Int J Hyperth. 2010;26(8):737–47.

9. Rappa F, Farina F, Zummo G, David S, Campanella C, Carini F, Tomasello G,Damiani P, Cappello F, EC DEM, et al. HSP-molecular chaperones in cancerbiogenesis and tumor therapy: an overview. Anticancer Res. 2012;32(12):5139–50.

10. Ciocca DR, Cayado-Gutierrez N, Maccioni M, Cuello-Carrion FD. Heat shockproteins (HSPs) based anti-cancer vaccines. Curr Mol Med. 2012;12(9):1183–97.

11. Cayado-Gutierrez N, Moncalero VL, Rosales EM, Beron W, Salvatierra EE,Alvarez-Olmedo D, Radrizzani M, Ciocca DR. Downregulation of Hsp27(HSPB1) in MCF-7 human breast cancer cells induces upregulation of PTEN.Cell Stress Chaperones. 2013;18(2):243–9.

12. Zhu Y, Qiu P, Ji Y. TCGA-assembler: open-source software for retrieving andprocessing TCGA data. Nat Methods. 2014;11(6):599–600.

13. Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, Speed D,Lynch AG, Samarajiwa S, Yuan Y, et al. The genomic and transcriptomicarchitecture of 2,000 breast tumours reveals novel subgroups. Nature. 2012;486(7403):346–52.

14. Ciriello G, Gatza ML, Beck AH, Wilkerson MD, Rhie SK, Pastore A, Zhang H,McLellan M, Yau C, Kandoth C, et al. Comprehensive molecular portraits ofinvasive lobular breast Cancer. Cell. 2015;163(2):506–19.

15. Yadav BS, Chanana P, Jhamb S. Biomarkers in triple negative breast cancer:a review. World J Clin Oncol. 2015;6(6):252–63.

16. Carvalho FM, Bacchi LM, Pincerato KM, Van de Rijn M, Bacchi CE. Geographicdifferences in the distribution of molecular subtypes of breast cancer in Brazil.BMC Womens Health. 2014;14:102.

17. Love MI, Huber W, Anders S. Moderated estimation of fold change anddispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550.

18. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package fordifferential expression analysis of digital gene expression data. Bioinformatics.2010;26(1):139–40.

19. Soneson C, Delorenzi M. A comparison of methods for differential expressionanalysis of RNA-seq data. BMC Bioinformatics. 2013;14:91.

20. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practicaland Powerful Approach to Multiple Testing. Journal of the Royal StatisticalSociety Series B (Methodological). 1995;57(1):289–300.

21. Bland JM, Altman DG. Statistical methods for assessing agreement betweentwo methods of clinical measurement. Lancet. 1986;1(8476):307–10.

22. Altman DG, McShane LM, Sauerbrei W, Taube SE. Reporting recommendationsfor tumor marker prognostic studies (REMARK): explanation and elaboration.BMC Med. 2012;10:51.

23. Jarosz D. Hsp90: a global regulator of the genotype-to-phenotype map incancers. Adv Cancer Res. 2016;129:225–47.

24. Calderwood SK, Khaleque MA, Sawyer DB, Ciocca DR. Heat shock proteins incancer: chaperones of tumorigenesis. Trends Biochem Sci. 2006;31(3):164–72.

25. Ciocca DR, Fanelli MA. Cuello-Carrión FD. In: Calderwood SK (eds.): implications ofheat shock proteins in carcinogenesis and Cancer progression: springerNetherlands; 2007.

26. Schulz R, Streller F, Scheel AH, Ruschoff J, Reinert MC, Dobbelstein M,Marchenko ND, Moll UM. HER2/ErbB2 activates HSF1 and thereby controlsHSP90 clients including MIF in HER2-overexpressing breast cancer. CellDeath Dis. 2014;5:e980.

27. Ozgur A, Tutar L, Tutar Y. Regulation of heat shock proteins by miRNAs inhuman breast cancer. Microrna. 2014;3(2):118–35.

28. Hadizadeh Esfahani A, Sverchkova A, Saez-Rodriguez J, Schuppert AA,Brehme M. A systematic atlas of chaperome deregulation topologiesacross the human cancer landscape. PLoS Comput Biol. 2018;14(1):e1005890.

29. Hartl FU, Bracher A, Hayer-Hartl M. Molecular chaperones in protein foldingand proteostasis. Nature. 2011;475(7356):324–32.

30. Seo S, Baye LM, Schulz NP, Beck JS, Zhang Q, Slusarski DC, Sheffield VC. BBS6,BBS10, and BBS12 form a complex with CCT/TRiC family chaperonins andmediate BBSome assembly. Proc Natl Acad Sci U S A. 2010;107(4):1488–93.

31. Guen VJ, Chavarria TE, Kroger C, Ye X, Weinberg RA, Lees JA. EMT programspromote basal mammary stem cell and tumor-initiating cell stemness byinducing primary ciliogenesis and hedgehog signaling. Proc Natl Acad Sci US A. 2017;114(49):E10532–9.

32. Legare S, Chabot C, Basik M. SPEN, A new player in primary cilia formationand cell migration in breast cancer. Breast Cancer Res 2017; 19(1):104.

33. Cui X, Hu ZP, Li Z, Gao PJ, Zhu JY. Overexpression of chaperonin containingTCP1, subunit 3 predicts poor prognosis in hepatocellular carcinoma. WorldJ Gastroenterol. 2015;21(28):8588–604.

34. Xiong Y, Wu S, Du Q, Wang A, Wang Z. Integrated analysis of gene expressionand genomic aberration data in osteosarcoma (OS). Cancer Gene Ther. 2015;22(11):524–9.

35. Penzvalto Z, Lanczky A, Lenart J, Meggyeshazi N, Krenacs T, Szoboszlai N,Denkert C, Pete I, Gyorffy B. MEK1 is associated with carboplatin resistanceand is a prognostic biomarker in epithelial ovarian cancer. BMC Cancer.2014;14:837.

36. Zhang Y, Wang Y, Wei Y, Wu J, Zhang P, Shen S, Saiyin H, Wumaier R, Yang X,Wang C, et al. Molecular chaperone CCT3 supports proper mitotic progressionand cell proliferation in hepatocellular carcinoma cells. Cancer Lett. 2016;372(1):101–9.

37. Nibbe RK, Markowitz S, Myeroff L, Ewing R, Chance MR. Discovery and scoringof protein interaction subnetworks discriminative of late stage human coloncancer. Mol Cell Proteomics. 2009;8(4):827–45.

38. Keenan J, Murphy L, Henry M, Meleady P, Clynes M. Proteomic analysis ofmultidrug-resistance mechanisms in adriamycin-resistant variants of DLKP,a squamous lung cancer cell line. Proteomics. 2009;9(6):1556–66.

39. Ooe A, Kato K, Noguchi S. Possible involvement of CCT5, RGS3, and YKT6genes up-regulated in p53-mutated tumors in resistance to docetaxel inhuman breast cancers. Breast Cancer Res Treat. 2007;101(3):305–15.

40. Guest ST, Kratche ZR, Bollig-Fischer A, Haddad R, Ethier SP. Two members ofthe TRiC chaperonin complex, CCT2 and TCP1 are essential for survival ofbreast cancer cells and are linked to driving oncogenes. Exp Cell Res. 2015;332(2):223–35.

41. Roh SH, Kasembeli M, Bakthavatsalam D, Chiu W, Tweardy DJ. Contributionof the type II chaperonin, TRiC/CCT, to oncogenesis. Int J Mol Sci. 2015;16(11):26706–20.

Zoppino et al. BMC Cancer (2018) 18:700 Page 15 of 17

42. Llorca O, Martin-Benito J, Ritco-Vonsovici M, Grantham J, Hynes GM, WillisonKR, Carrascosa JL, Valpuesta JM. Eukaryotic chaperonin CCT stabilizes actinand tubulin folding intermediates in open quasi-native conformations.EMBO J. 2000;19(22):5971–9.

43. Trinidad AG, Muller PA, Cuellar J, Klejnot M, Nobis M, Valpuesta JM, VousdenKH. Interaction of p53 with the CCT complex promotes protein folding andwild-type p53 activity. Mol Cell. 2013;50(6):805–17.

44. Kasembeli M, Lau WC, Roh SH, Eckols TK, Frydman J, Chiu W, Tweardy DJ.Modulation of STAT3 folding and function by TRiC/CCT chaperonin. PLoSBiol. 2014;12(4):e1001844.

45. Teshome M, Hunt KK. Neoadjuvant therapy in the treatment of breast cancer.Surg Oncol Clin N Am. 2014;23(3):505–23.

46. Coumans JV, Gau D, Poljak A, Wasinger V, Roy P, Moens PD. Profilin-1overexpression in MDA-MB-231 breast cancer cells is associated with alterationsin proteomics biomarkers of cell proliferation, survival, and motility as revealed byglobal proteomics analyses. OMICS. 2014;18(12):778–91.

47. Czarnecka AM, Campanella C, Zummo G, Cappello F. Heat shock protein 10and signal transduction: a "capsula eburnea" of carcinogenesis? Cell StressChaperones. 2006;11(4):287–94.

48. Mayer MP, Bukau B. Hsp70 chaperones: cellular functions and molecularmechanism. Cell Mol Life Sci. 2005;62(6):670–84.

49. Yang Z, Zhuang L, Szatmary P, Wen L, Sun H, Lu Y, Xu Q, Chen X. Upregulationof heat shock proteins (HSPA12A, HSP90B1, HSPA4, HSPA5 and HSPA6) intumour tissues is associated with poor outcomes from HBV-related early-stagehepatocellular carcinoma. Int J Med Sci. 2015;12(3):256–63.

50. Asahi H, Koshida K, Hori O, Ogawa S, Namiki M. Immunohistochemicaldetection of the 150-kDa oxygen-regulated protein in bladder cancer. BJUInt. 2002;90(4):462–6.

51. Miyagi T, Hori O, Koshida K, Egawa M, Kato H, Kitagawa Y, Ozawa K, Ogawa S,Namiki M. Antitumor effect of reduction of 150-kDa oxygen-regulated proteinexpression on human prostate cancer cells. Int J Urol. 2002;9(10):577–85.

52. Ozawa K, Tsukamoto Y, Hori O, Kitao Y, Yanagi H, Stern DM, Ogawa S. Regulationof tumor angiogenesis by oxygen-regulated protein 150, an inducibleendoplasmic reticulum chaperone. Cancer Res. 2001;61(10):4206–13.

53. Tsukamoto Y, Kuwabara K, Hirota S, Kawano K, Yoshikawa K, Ozawa K,Kobayashi T, Yanagi H, Stern DM, Tohyama M, et al. Expression of the 150-kd oxygen-regulated protein in human breast cancer. Lab Investig. 1998;78(6):699–706.

54. Cerezo M, Rocchi S. New anti-cancer molecules targeting HSPA5/BIP toinduce endoplasmic reticulum stress, autophagy and apoptosis. Autophagy.2017;13(1):216–7.

55. Wang J, Lee J, Liem D, Ping P. HSPA5 gene encoding Hsp70 chaperone BiPin the endoplasmic reticulum. Gene. 2017;618:14–23.

56. Uckun FM, Qazi S, Ozer Z, Garner AL, Pitt J, Ma H, Janda KD. Inducing apoptosisin chemotherapy-resistant B-lineage acute lymphoblastic leukaemia cells bytargeting HSPA5, a master regulator of the anti-apoptotic unfolded proteinresponse signalling network. Br J Haematol. 2011;153(6):741–52.

57. Tanaka M, Mun S, Harada A, Ohkawa Y, Inagaki A, Sano S, Takahashi K, Izumi Y,Osada-Oka M, Wanibuchi H, et al. Hsc70 contributes to cancer cell survival bypreventing Rab1A degradation under stress conditions. PLoS One. 2014;9(5):e96785.

58. Zuiderweg ER, Hightower LE, Gestwicki JE. The remarkable multivalency ofthe Hsp70 chaperones. Cell Stress Chaperones. 2017;22(2):173–89.

59. Choi J, Djebbar S, Fournier A, Labrie C. The co-chaperone DNAJC12 binds toHsc70 and is upregulated by endoplasmic reticulum stress. Cell StressChaperones. 2014;19(3):439–46.

60. Gundersen CB, Kohan SA, Souda P, Whitelegge JP, Umbach JA. Cysteinestring protein beta is prominently associated with nerve terminals andsecretory organelles in mouse brain. Brain Res. 2010;1332:1–11.

61. Mirzaei MR, Asadi M, Mowla SJ, Hassanshahi G, Ahmadi Z. Down-regulationof HSP40 gene family following OCT4B1 suppression in human tumor celllines. Iran J Basic Med Sci. 2016;19(2):187–93.

62. Xie J, Marusich MF, Souda P, Whitelegge J, Capaldi RA. The mitochondrialinner membrane protein mitofilin exists as a complex with SAM50, metaxins1 and 2, coiled-coil-helix coiled-coil-helix domain-containing protein 3 and6 and DnaJC11. FEBS Lett. 2007;581(18):3545–9.

63. Ioakeimidis F, Ott C, Kozjak-Pavlovic V, Violitzi F, Rinotas V, Makrinou E,Eliopoulos E, Fasseas C, Kollias G, Douni E. A splicing mutation in thenovel mitochondrial protein DNAJC11 causes motor neuron pathologyassociated with cristae disorganization, and lymphoid abnormalities inmice. PLoS One. 2014;9(8):e104237.

64. Pan Z, Chen S, Pan X, Wang Z, Han H, Zheng W, Wang X, Li F, Qu S, Shao R.Differential gene expression identified in Uigur women cervical squamouscell carcinoma by suppression subtractive hybridization. Neoplasma. 2010;57(2):123–8.

65. Abu-Farha M, Cherian P, Al-Khairi I, Tiss A, Khadir A, Kavalakatt S, Warsame S,Dehbi M, Behbehani K, Abubaker J. DNAJB3/HSP-40 cochaperone improvesinsulin signaling and enhances glucose uptake in vitro through JNK repression.Sci Rep. 2015;5:14448.

66. Katsogiannou M, Andrieu C, Rocchi P. Heat shock protein 27 phosphorylationstate is associated with cancer progression. Front Genet. 2014;5:346.

67. Fanelli MA, Montt-Guevara M, Diblasi AM, Gago FE, Tello O, Cuello-CarrionFD, Callegari E, Bausero MA, Ciocca DR. P-cadherin and beta-catenin areuseful prognostic markers in breast cancer patients; beta-catenin interactswith heat shock protein Hsp27. Cell Stress Chaperones. 2008;13(2):207–20.

68. Simoes-Correia J, Silva DI, Melo S, Figueiredo J, Caldeira J, Pinto MT, Girao H,Pereira P, Seruca R. DNAJB4 molecular chaperone distinguishes WT frommutant E-cadherin, determining their fate in vitro and in vivo. Hum MolGenet. 2014;23(8):2094–105.

69. Guey LT, Garcia-Closas M, Murta-Nascimento C, Lloreta J, Palencia L,Kogevinas M, Rothman N, Vellalta G, Calle ML, Marenne G, et al.Genetic susceptibility to distinct bladder cancer subphenotypes. EurUrol. 2010;57(2):283–92.

70. Su Q, Wang Y, Zhao J, Ma C, Wu T, Jin T, Xu J. Polymorphisms of PRLHR andHSPA12A and risk of gastric and colorectal cancer in the Chinese Hanpopulation. BMC Gastroenterol. 2015;15:107.

71. Salhia B, Kiefer J, Ross JT, Metapally R, Martinez RA, Johnson KN, DiPerna DM,Paquette KM, Jung S, Nasser S, et al. Integrated genomic and epigenomicanalysis of breast cancer brain metastasis. PLoS One. 2014;9(1):e85448.

72. Arrigo AP, Simon S, Gibert B, Kretz-Remy C, Nivon M, Czekalla A, Guillet D,Moulin M, Diaz-Latoud C, Vicart P. Hsp27 (HspB1) and alphaB-crystallin(HspB5) as therapeutic targets. FEBS Lett. 2007;581(19):3665–74.

73. Kim HS, Lee Y, Lim YA, Kang HJ, Kim LS. alphaB-Crystallin is a novel Oncoproteinassociated with poor prognosis in breast Cancer. J Breast Cancer. 2011;14(1):14–9.

74. Ivanov O, Chen F, Wiley EL, Keswani A, Diaz LK, Memmel HC, Rademaker A,Gradishar WJ, Morrow M, Khan SA, et al. alphaB-crystallin is a novel predictor ofresistance to neoadjuvant chemotherapy in breast cancer. Breast Cancer ResTreat. 2008;111(3):411–7.

75. Yilmaz M, Karatas OF, Yuceturk B, Dag H, Yener M, Ozen M. Alpha-B-crystallinexpression in human laryngeal squamous cell carcinoma tissues. Head Neck.2015;37(9):1344–8.

76. Boelens WC. Role of small heat shock protein HspB5 in Cancer. The BigBook on Small Heat Shock Proteins Springer. 2015;

77. Lung HL, Lo CC, Wong CC, Cheung AK, Cheong KF, Wong N, Kwong FM,Chan KC, Law EW, Tsao SW, et al. Identification of tumor suppressive activityby irradiation microcell-mediated chromosome transfer and involvement ofalpha B-crystallin in nasopharyngeal carcinoma. Int J Cancer. 2008;122(6):1288–96.

78. Huang Z, Cheng Y, Chiu PM, Cheung FM, Nicholls JM, Kwong DL, Lee AW,Zabarovsky ER, Stanbridge EJ, Lung HL, et al. Tumor suppressor alpha B-crystallin (CRYAB) associates with the cadherin/catenin adherens junctionand impairs NPC progression-associated properties. Oncogene. 2012;31(32):3709–20.

79. van de Schootbrugge C, Bussink J, Span PN, Sweep FC, Grenman R, Stegeman H,Pruijn GJ, Kaanders JH, Boelens WC. alphaB-crystallin stimulates VEGF secretionand tumor cell migration and correlates with enhanced distant metastasis inhead and neck squamous cell carcinoma. BMC Cancer. 2013;13:128.

80. Koga Y, Pelizzola M, Cheng E, Krauthammer M, Sznol M, Ariyan S, Narayan D,Molinaro AM, Halaban R, Weissman SM. Genome-wide screen of promotermethylation identifies novel markers in melanoma. Genome Res. 2009;19(8):1462–70.

81. Noda T, Kumada T, Takai S, Matsushima-Nishiwaki R, Yoshimi N, Yasuda E,Kato K, Toyoda H, Kaneoka Y, Yamaguchi A, et al. Expression levels of heatshock protein 20 decrease in parallel with tumor progression in patientswith hepatocellular carcinoma. Oncol Rep. 2007;17(6):1309–14.

82. Matsushima-Nishiwaki R, Toyoda H, Nagasawa T, Yasuda E, Chiba N, OkudaS, Maeda A, Kaneoka Y, Kumada T, Kozawa O. Phosphorylated heat shockprotein 20 (HSPB6) regulates transforming growth factor-alpha-inducedmigration and invasion of hepatocellular carcinoma cells. PLoS One. 2016;11(4):e0151907.

83. Sudnitsyna M, Sluchanko N, Gusev N. HSPB6 (Hsp20) as a Versatile MolecularRegulator. The Big Book on Small Heat Shock Proteins Springer. 2015.

Zoppino et al. BMC Cancer (2018) 18:700 Page 16 of 17

84. Ju YT, Kwag SJ, Park HJ, Jung EJ, Jeong CY, Jeong SH, Lee YJ, Choi SK, KangKR, Hah YS, et al. Decreased expression of heat shock protein 20 incolorectal cancer and its implication in tumorigenesis. J Cell Biochem. 2015;116(2):277–86.

85. Lin J, Deng Z, Tanikawa C, Shuin T, Miki T, Matsuda K, Nakamura Y.Downregulation of the tumor suppressor HSPB7, involved in the p53pathway, in renal cell carcinoma by hypermethylation. Int J Oncol.2014;44(5):1490–8.

86. Aran D, Sirota M, Butte AJ. Systematic pan-cancer analysis of tumour purity.Nat Commun. 2015;6:8971.

87. Wu J, Liu T, Rios Z, Mei Q, Lin X, Cao S. Heat shock proteins and Cancer.Trends Pharmacol Sci. 2017;38(3):226–56.

Zoppino et al. BMC Cancer (2018) 18:700 Page 17 of 17